Category Archives: Latent GOLD®

Course: Latent Class Models for Multilevel and Longitudinal Data

Date:
July 8-10, 2015

Teacher(s):
Jeroen K. Vermunt, Tilburg University
Margot Sijssens-Bennink, Statistical Innovations

Location:
Barcelona , Spain

Course content:
This course deals with various more advanced application types of latent class (LC) analysis. These concern applications with multilevel and longitudinal data sets. More specifically, you will learn how to use LC regression models, LC growth models, latent Markov models, and multilevel LC models. During this could we will use the Latent GOLD computer program, including the Syntax module.

First we will look into the data organization for these more advanced LC analysis applications. These are univariate or multivariate two-level data files, which contain multiple records per unit, with a single or multiple response variables and in most cases also predictor variables.

The LC or mixture regression model is a two-level regression model in which the intercept and predictor effects are allowed to vary across units by assuming that units belong to latent classes. This model can also be seen as a semi-parametric variant of the standard random effects regression model. Application types include standard multilevel regression analysis and the modeling of data from rating or choice-based conjoint experiments.

One special type of application of the LC regression model involves its use with longitudinal data sets. Note that these are also two-level data sets, with time variables as the main predictors. Such models are referred to as LC growth or LC trajectory models. LC growth models may contain continuous random effects in addition to latent classes.

The latent Markov or latent transition model is an extension of the simple LC cluster model for use with longitudinal data. Latent Markov models differ from LC growth models in that they can also be used with multiple responses and in that respondents are allowed to change their class membership over time. Basically, we model the transition across latent states. Extension of the basic model include models with covariates and models with a mover-stayer structure.

The multilevel LC model is an extension of the simple LC cluster model to the situation in which lower-level units are nested within higher-level units. In this model not only individuals are clustered into a small number of classes, but also the groups are clustered into group-level classes.

Requirement for successful participation in this course are basic knowledge of LC analysis (for example, the introductory course of the summer school), multilevel regression analysis, and longitudinal data analysis.

More information and registration:
http://www.upf.edu/survey/Summer/

Course: Introduction to Latent Class Cluster Analysis

Date:
July 6-8, 2015

Teacher(s):
Jeroen K. Vermunt, Tilburg University
Margot Sijssens-Bennink, Statistical Innovations

Location:
Barcelona , Spain

Course content:
Latent class (LC) analysis is used in a broad range of research fields with the aim to cluster respondents into a small number subgroups called latent classes. Typically the observed variables used in a LC analysis are categorical variables, but it also possible to use continuous variables or counts. In this introductory course you will learn how to perform such an analysis by yourself.  We will not only focus on practical skills, but also discuss the most important parts of the underlying theory. All steps will be illustrated using empirical examples. During this course we will use the software package Latent GOLD.

First, using a simple example application, we present the structure of the basic LC model for categorical variables and discuss its key model assumptions. Subsequently, we switch to a more realistic example to discuss the important issue of model selection or, more specifically, how to determine the number of latent classes. For model selection, we use information criteria (AIC, AIC3, and BIC), goodness-of-fit statistics (possibly with bootstrap p-values), bivariate residuals, bootstrapped likelihood-ratio tests, and Wald tests.

Once a model with a specific number of latent classes is selected, respondents are classified into the latent classes based on their posterior class membership probabilities. We discuss how these classifications are obtained, as well as how the quality of the classifications can be evaluated using so-called classification statistics.

Subsequently, we will discuss various important extensions of the basic LC model, such as LC models which relax the local dependence assumption, and LC models for ordinal, continuous, and count variables.

Finally, we deal with the important issue on how to investigate the relationship between the individuals’ class memberships with covariates and distal outcomes. This can be done in various ways, among others using the recently developed bias adjusted three-step approach. We also discuss multiple-group LC analysis.

You don’t need any background in LC analysis to be able to follow this introductory course. Also researchers with some experience in LC analysis will benefit from reviewing the basics since they will pick up more of the details while beginning users can focus on the global learning goals of the course.

More information and registration:
http://www.upf.edu/survey/Summer/